Measure and troubleshoot

Search analytics

Queries, click-through, conversion, zero results.

Search analytics is the only place where buyers tell you, in their own words, what they wanted. Four reports carry almost all of the value, and one number tells you whether search is working at all. This page defines them, so that the number you track from a spreadsheet this quarter is the same number the Cockpit shows later.

Cockpit screens: planned. The Search app's analytics screens and Analytics Studio are not yet on your tenant. Today the four reports come from your storefront's search log and tracking, as an export. Before you read anything into the numbers, confirm that search events fire for signed-in buyers, on mobile, and inside a punchout session; analytics counts what the storefront reports, so it is only as complete as your tracking. That check stays true once the screens ship.

The four reports

ReportWhat it showsWhat you do with it
Top queriesMost frequent searches, with result countsCheck the first three results by hand. These are your shop windows.
Zero resultsSearches returning nothingThe weekly loop
Low click-throughSearches with results that nobody clicksThe results are wrong rather than missing, which is the harder problem
Search-to-orderSearches that ended in an orderThe KPI, below

Low click-through is the report people skip and should not. A query with 200 results and a two percent click rate is worse than a zero-result query: the buyer got a wall of near-misses, could not tell them apart, and gave up. The usual causes are a facet panel that does not fit the category, results that all look identical because the differentiating attribute is not shown in the list, or a boost that promoted the wrong family.

Search-to-order rate: the KPI

One number, defined once, tracked forever:

Of the sessions in which a buyer used search, what share ended in an order?

Compare it against the same number for sessions that did not use search. In a healthy B2B shop, searchers convert better than browsers, because they arrived knowing what they wanted. If your searchers convert worse than your browsers, search is costing you orders and everything in this guide is worth doing this month.

Two secondary numbers belong next to it:

  • Zero-result rate. The share of searches returning nothing. As a working target, under three percent is good; over ten percent means the vocabulary work has never started.
  • Refinement rate. The share of searches followed by another search within the same session. High refinement means buyers did not get it right first time, which is usually a ranking problem rather than a data problem.
Do not track average result count as a quality metric. It rewards the wrong thing. A search that returns one correct article is better than one that returns sixty, and a configuration tuned to raise the average is a configuration tuned to make search useless.

Read every report per channel and per locale

Aggregated search analytics hides the problems that matter. The index is scoped by channel and locale, so the split is natural. Read every report by:

  • Channel. Your web shop and a punchout catalog have different assortments, different buyers and different vocabulary. A zero-result query in a punchout catalog is usually an assortment question: the article exists and is not assigned to that channel. See Channels, locales and completeness.
  • Locale. The English zero-result rate is almost always worse than the German one, because the catalog was translated and the vocabulary was not. This split tells you when a market needs translation work rather than search work. A product with no English name is not found in the English storefront at all; see Common search problems.
  • Signed in or not. Anonymous visitors search differently: more browsing, fewer article numbers. If you sell only to signed-in customers, the anonymous numbers are noise; separate them so they do not drag the KPI around.

The monthly review

Thirty minutes, same day each month, from the export until the screens exist and from the screens afterwards:

  1. Read the search-to-order rate per channel against last month. Explain any move of more than a couple of points before doing anything else.
  2. Read the top 20 queries and check the first three results for each. This is manual, and it is the highest-value half hour in the whole area.
  3. Read the low click-through list and pick the worst three. For each, decide whether it is a facet problem, a ranking problem or a data problem.
  4. Confirm the zero-result rate is falling, and that last month's fixes did not come back.
  5. Once boosts and pins exist, count them. A count that grows every month means ranking is being patched query by query.

Feeding it back

Search analytics is more than a search tool.

  • Purchasing gets the "genuinely not sold" list: real demand for articles you do not stock.
  • The catalog team gets the attribute gaps, which improve exports and classification as well as search.
  • Sales gets the assortment gaps: articles buyers looked for inside a channel they are not entitled to.
  • Marketing gets the vocabulary your customers use, which is usually not the vocabulary on your category pages.

When Analytics Studio ships, the search-to-order rate belongs next to the order and customer metrics in Metrics that matter. Until then, keep it in the same spreadsheet as your monthly order numbers, so the two are read together.

Next